Machine Learning Analysis of State of Polarization Changes to Detect Optical Fiber Tampering

Licentiate thesis, 2024

Fiber optic networks are the backbone of modern communications, supporting a vast range of services, from internet connectivity to critical infrastructure operations, such as defense, healthcare, and finance. Their ability to transmit data at ultra-high rates over long distances with minimal loss makes them the preferred medium for secure and efficient data transmission. However,  fiber optic installations face various security and physical damage threats which can compromise the reliability, integrity, and confidentiality of the transmitted data. The threats range from physical damage caused by accidental fiber cuts or mechanical vibrations to sophisticated eavesdropping attacks that exploit the physical properties of optical fibers to gain unauthorized access to the transmitted data.
Given the critical role of fiber optic networks in today's interconnected world, ensuring their security, reliability, and resilience is paramount. Effective monitoring is a key aspect of maintaining network security, as it enables the early detection of potential threats and disturbances. Traditional monitoring systems are often limited in scope, costly, and struggle to detect more subtle disturbances like unauthorized tapping or eavesdropping. Recent advances in Machine Learning (ML) offer new avenues for enhancing the detection and diagnostics of anomalies in optical networks.
This thesis investigates the use of the State of Polarization (SOP) of light within optical fibers as a novel technique for monitoring environmental changes and detecting security threats. By employing a range of ML techniques, including supervised, semi-supervised, and unsupervised learning, this research aims at identifying and classifying disturbances that may indicate mechanical damage or security breaches. The work presented in this thesis demonstrates how SOP analysis, enhanced by advanced ML models, can improve the detection capabilities of fiber optic cables as sensing devices, providing a cost-effective and scalable solution for safeguarding data integrity, confidentiality, and network availability. The findings of this research contribute to the development of intelligent and adaptive security systems for fiber optic infrastructure.